Humidity Monitoring Using a Flexible Polymer- based Microwave Sensor and Machine Learning.

This work presents humidity monitoring using a highly sensitive flexible microwave sensor associated with polyethyleneimine sensitive material with high endurance against temperature by a machine learning approach. A climatic chamber was used to generate humidity at different temperatures and a commercialized humidity and temperature sensor was used as a reference. The sensor showed a high frequency sensitivity (-3.65 and -7.69 MHz/%RH in a range of 30 - 50 %RH and 50 - 70%RH respectively), low hysteresis, good reversibility and repeatability. Moreover, the extracted sensing features were associated to linear regression, support vector machine, random forest and k-nearest neighbours regression algorithms for humidity prediction. The performance of the different models was evaluated and random forest (MAE: 1.63 %RH, R-2: 0.970, pred time: 0.44s) and k-nearest neighbours ((MAE: 1.52 %RH, R-2: 0.971, pred time: 0.12s) showed the best results on prediction on the test data set.